2.5 Semisimple Lie Algebras
2.5.2 Root Space Decomposition
g=M
λ
gλ(X),
andSacts byλongλ(X). TakingD=adX,Y∈gλ(X),Z∈gµ(X), andksufficiently large in (2.45), we see that
[gλ(X),gµ(X)]⊂gλ+µ(X). (2.46) HenceSis a derivation ofg. By Corollary 2.5.12 there existsXs∈gsuch that adXs=
S. SetXn=X−Xs. ut
2.5 Semisimple Lie Algebras 115
XY v=Y X v+ [X,Y]v=0
for allX∈h, since[X,Y]∈h. ThusW is invariant underY, so there exists a nonzero vectorv0∈W such thatY v0=0. It follows thatgv0=0. ut Corollary 2.5.15.There exists a basis for V in which the elements ofgare repre- sented by strictly upper-triangular matrices.
Proof. This follows by repeated application of Theorem 2.5.14, replacingV by
V/Cv0at each step. ut
Corollary 2.5.16.Suppose g is a semisimple Lie algebra. Then there exists a nonzero element X∈gsuch thatadX is semisimple.
Proof. We argue by contradiction. Ifgcontained no nonzero elementsX with adX semisimple, then Corollary 2.5.13 would imply that adX is nilpotent for allX∈g.
Hence Corollary 2.5.15 would furnish a basis forgsuch that adX is strictly upper triangular. But then adXadY would also be strictly upper triangular for allX,Y∈g, and hence the Killing form would be zero, contradicting Theorem 2.5.11. ut For the rest of this section we letgbe a semisimple Lie algebra. We call a subal- gebrah⊂gatoral subalgebraif adXis semisimple for allX∈h. Corollary 2.5.16 implies the existence of nonzero toral subalgebras.
Lemma 2.5.17.Lethbe a toral subalgebra. Then[h,h] =0.
Proof. LetX∈h. Thenhis an invariant subspace for the semisimple transformation adX. If[X,h]6=0 then there would exist an eigenvalueλ 6=0 and an eigenvector Y ∈hsuch that[X,Y] =λY. But then
(adY)(X) =−λY6=0, (adY)2(X) =0,
which would imply that adY is not a semisimple transformation. Hence we must
have[X,h] =0 for allX∈h. ut
We shall call a toral subalgebrah⊂gaCartan subalgebraif it has maximal di- mension among all toral subalgebras ofg. From Corollary 2.5.16 and Lemma 2.5.17 we see thatgcontains nonzero Cartan subalgebras and that Cartan subalgebras are abelian. We fix a choice of a Cartan subalgebrah. Forλ∈h∗let
gλ={Y∈g:[X,Y] =hλ,XiY for all X∈h}.
In particular,g0={Y ∈g: [X,Y] =0 for all X∈h}is thecentralizerofhing.
LetΦ ⊂g∗\ {0}be the set ofλ such thatgλ 6=0. We callΦ the set ofrootsofh ong. Since the mutually commuting linear transformations adXare semisimple (for X∈h), there is aroot space decomposition
g=g0⊕M
λ∈Φ
gλ .
LetBdenote the Killing form ofg. By the same arguments used for the classical groups in Sections 2.4.1 and 2.4.2 (but now usingBinstead of the trace form on the defining representation of a classical group), it follows that
1. [gλ,gµ]⊂gλ+µ ;
2. B(gλ,gµ) =0 ifλ+µ6=0 ;
3. the restriction ofBtog0×g0is nondegenerate;
4. ifλ∈Φthen−λ∈Φand the restriction ofBtogλ×g−λ is nondegenerate.
New arguments are needed to prove the following key result:
Proposition 2.5.18.A Cartan algebra is its own centralizer ing; thush=g0. Proof. Sincehis abelian, we haveh⊂g0. LetX∈g0and letX=Xs+Xnbe the Jordan decomposition ofXgiven by Corollary 2.5.13.
(i) XsandXnare ing0.
Indeed, since [X,h] =0 and the adjoint representation of g is faithful, we have [Xs,h] =0. HenceXs∈hby the maximality ofh, which implies thatXn=X−Xsis also inh.
(ii) The restriction ofBtoh×his nondegenerate.
To prove this, let 06=h∈h. Then by property (3) there exists X ∈g0 such that B(h,X)6=0. SinceXn∈g0by (i), we have[h,Xn] =0 and hence adhadXnis nilpotent ong. ThusB(h,Xn) =0 and soB(h,Xs)6=0. SinceXs∈h, this proves (ii).
(iii) [g0,g0] =0 .
For the proof of (iii), we observe that ifX∈g0, then adXsacts by zero ong0, since Xs∈h. Hence adX|g0=adXn|g0 is nilpotent. Suppose for the sake of contradiction that [g0,g0]6=0 and consider the adjoint action of g0 on the invariant subspace [g0,g0]. By Theorem 2.5.14 there would exist 06=Z∈[g0,g0]such that[g0,Z] =0.
Then[g0,Zn] =0 and hence adYadZnis nilpotent for allY ∈g0. This implies that B(Y,Zn) =0 for allY ∈g0, so we conclude from (3) thatZn=0. ThusZ=Zsmust be inh. Now
B(h,[X,Y]) =B([h,X],Y) =0 for all h∈h and X,Y ∈g0. Henceh∩[g0,g0] =0 by (ii), and soZ=0, giving a contradiction.
It is now easy to complete the proof of the proposition. IfX,Y∈g0then adXnadY is nilpotent, sinceg0is abelian by (iii). HenceB(Xn,Y) =0, and soXn=0 by (3).
ThusX=Xs∈h. ut
Corollary 2.5.19.Let gbe a semisimple Lie algebra andha Cartan subalgebra.
Then
g=h⊕M
λ∈Φ
gλ. (2.47)
Hence if Y ∈gand [Y,h]⊂h, then Y ∈h. In particular,his a maximal abelian subalgebra ofg.
2.5 Semisimple Lie Algebras 117
Since the formBis nondegenerate onh×h, it defines a bilinear form onh∗that we denote by(α,β).
Theorem 2.5.20.The roots and root spaces satisfy the following properties:
1.Φ spansh∗.
2. Ifα ∈Φ thendim[gα,g−α] =1and there is a unique element hα ∈[gα,g−α] such thathα,hαi=2(call hα thecoroottoα).
3. Ifα∈Φ and c∈Cthen cα∈Φif and only if c=±1. Alsodimgα=1.
4. Letα,β∈Φ withβ6=±α. Let p be the largest integer j≥0withβ+jα∈Φ and let q be the largest integer k≥0withβ−kα∈Φ. Then
hβ,hαi=q−p∈Z (2.48)
andβ+rα∈Φ for all integers r with−q≤r≤p. Henceβ− hβ,hαiα∈Φ. 5. Ifα,β ∈Φ andα+β∈Φ, then[gα,gβ] =gα+β .
Proof. (1): Ifh∈handhα,hi=0 for allα∈Φ, then[h,gα] =0 and hence[h,g] =0.
The center ofgis trivial, sinceghas no abelian ideals, soh=0. ThusΦspansh∗. (2): LetX∈gαandY∈g−α. Then[X,Y]∈g0=hand forh∈hwe have
B(h,[X,Y]) =B([h,X],Y) =hα,hiB(X,Y).
Thus[X,Y]corresponds toB(X,Y)α under the isomorphismh∼=h∗given by the formB. SinceBis nondegenerate ongα×g−α, it follows that dim[gα,g−α] =1.
SupposeB(X,Y)6=0 and setH= [X,Y]. Then 06=H∈h. Ifhα,Hi=0 thenH would commute withXandY, and hence adHwould be nilpotent by Lemma 2.5.1, which is a contradiction. Hencehα,Hi 6=0 and we can rescaleX andY to obtain elementseα∈gα andfα∈g−α such thathα,hαi=2, wherehα= [eα,fα].
(3): Lets(α) =Span{eα,fα,hα} ∼=sl(2,C)and set Mα=Chα+
∑
c6=0
gcα.
Since[eα,gcα]⊂g(c+1)α, [fα,gcα]⊂g(c−1)α, and[eα,g−α] = [fα,gα] =Chα, we see thatMα is invariant under the adjoint action ofs(α).
The eigenvalues of adhαonMα are 2cwith multiplicity dimgcαand 0 with mul- tiplicity one. By the complete reducibility of representations ofsl(2,C)(Theorem 2.3.6) and the classification of irreducible representations (Proposition 2.3.3) these eigenvalues must be integers. Hencecα∈Φimplies that 2cis an integer. The eigen- values in any irreducible representation are all even or all odd. Hencecα is not a root for any integercwith|c|>1, sinces(α)contains the zero eigenspace inMα. This also proves that the only irreducible component ofMα with even eigenvalues iss(α), and it occurs with multiplicity one.
Suppose(p+1/2)α ∈Φ for some positive integer p. Then adhα would have eigenvalues 2p+1,2p−1, . . . ,3,1 onMα, and hence(1/2)α would be a root. But
thenα could not be a root, by the argument just given, which is a contradiction.
Thus we conclude that Mα=Chα+Ceα+Cfα. Hence dimgα=1.
(4): The notion ofα root string throughβ from Section 2.4.2 carries over verba- tim, as does Lemma 2.4.3. Hence the argument in Corollary 2.4.5 applies.
(5): This follows from the same argument as Corollary 2.4.4. ut